Unitask Agent

Start finishing tasks instead of just organizing them: connect any AI agent to Unitask (unitask.app) to manage and do your tasks with secure prioritization, tags, time blocks and more.

Unitask is now in public beta. Anyone can sign up at https://unitask.app and create a scoped API token. Recommended runtimes: OpenClaw, Codex IDE, Claude Code, and any MCP-compatible client.

What it does

• Read tasks with token-efficient retrieval controls (fields, compact, description shaping)
• Create/update tasks (including subtasks via parent_id)
• Move subtasks between parent tasks
• Merge parent tasks (with preview first)
• Create/edit/delete tags
• Add/remove tags on tasks
• Soft-delete tasks
• Time-block your day (writes scheduled_start + duration_minutes)
• Read/update settings for onboarding and workflow preferences

What it helps you achieve

• See only what matters right now (today, upcoming, recurring, high-priority slices)
• Cut token usage for agent workflows by returning smaller task payloads first
• Turn high-level intent into concrete actions (create, re-prioritize, schedule, close loops)
• Keep change risk low with dry-run-first for move/merge operations

What it does not do yet

• No single-call bulk status update endpoint (agents should iterate task IDs)
• No undo endpoint for destructive operations
• No native "explain why this priority was chosen" inference endpoint (agents apply your own policy)

Setup

1. Sign up at https://unitask.app (public beta) if you do not already have an account.
2. In Unitask: Dashboard -> Settings -> API create an API token with scopes you need.
3. In your AI app / MCP client config, store token as UNITASK_API_KEY (never paste tokens in chat).
4. Optional but recommended one-time setup: set account timezone via update_settings({ settings: { timeZone: "<IANA>" } }).

Hosted MCP endpoint

• URL: https://unitask.app/api/mcp
• Header: Authorization: Bearer <UNITASK_API_KEY>

First 5 prompts

1. "What are my tasks for today, including overdue, sorted by priority? Use compact output."
2. "Show my upcoming tasks for the next 7 days using fields id,title,priority,due_date."
3. "Create a task: Draft beta launch update, due tomorrow at 5pm. Then tag it launch."
4. "Move subtask <task_id> under parent <parent_id>; preview first, then apply."
5. "Merge parent <source_parent_id> into <target_parent_id>; show dry-run summary before apply."

list_tasks quick controls

• Views/time windows: view, tz (optional override), window_days
• Date filters: due_from, due_to, start_from, start_to
• Targeted selectors: recurring_only, priority_buckets, has_scheduled, has_start_date, has_due_date
• Response shaping: fields, compact=compact|minimal, include_description, description_max_chars, description_ellipsis

If tz is omitted, Unitask defaults to account timezone from settings.timeZone.

For full tool contracts, see:

• SKILL.md (complete MCP tool list + scope/safety rules)
• USE_CASES.md (practical scenario matrix)
• EXAMPLES.md (top intents with prompt + expected output shape)

Examples Gallery

Prompt + expected-output gallery for top intents:

• EXAMPLES.md

Client configuration examples

Claude Code (remote HTTP MCP)

Create/edit ~/.claude.json:

{
  "mcpServers": {
    "unitask": {
      "type": "http",
      "url": "https://unitask.app/api/mcp",
      "headers": {
        "Authorization": "Bearer ${UNITASK_API_KEY}"
      }
    }
  }
}

VS Code (remote HTTP MCP)

Create .vscode/mcp.json in your workspace:

{
  "inputs": [
    {
      "id": "unitask_api_key",
      "type": "promptString",
      "description": "Unitask API key",
      "password": true
    }
  ],
  "servers": {
    "unitask": {
      "type": "http",
      "url": "https://unitask.app/api/mcp",
      "headers": {
        "Authorization": "Bearer ${input:unitask_api_key}"
      }
    }
  }
}

Safety

• Use the smallest scope needed.
• Scope guidance: read for retrieval, write for create/update, delete only when required.
• Keep dry_run=true first for compound actions (move/merge/timeblock apply).
• Confirm deletes unless user explicitly requested deletion.
• Avoid over-fetching: start with compact/projection queries and expand only for tasks you will act on.